Scaling Personalized Teaching with Pensive

## Overview Instructor: Dr. Tony Dear, Senior Lecturer, Department of Computer Science, Columbia University Tony Dear teaches large, computation-heavy courses at Columbia University, often with hundreds of students and multiple teaching assistants (TAs). Starting from Spring 2025, his team adopted Pensive, an AI-powered grading and tutoring platform, to enhance efficiency, speed and feedback quality for his high-enrollment courses Over the course of three terms (Spring, Summer, Fall 2025), Tony used Pensive on the following courses: - COMS W3203: Discrete Mathematics (~250 students) - COMS W4701: Artificial Intelligence (~50 students) ## Challenge Before using Pensive, grading and feedback were major bottlenecks for Tony’s classes. Providing written, **individualized** feedback for courses with more than 300 students was virtually impossible. - TAs spent half or more of their time commitment grading, an important task but one that drew time away from office hours and more direct student engagement. - Students received limited qualitative feedback beyond rubric checkmarks. - In-class engagement was uneven — extroverted students thrived, while quieter students hesitated to ask for help. Wanting to provide better feedback, and reduce the load on his TAs, Tony sought a tool that could streamline grading workflows, provide **personalized, high-quality feedback**, and create new avenues for interactive learning. ## Solution Tony and his team implemented both key features of Pensive: AI-Powered Grading and AI Tutoring. Both features created positive outcomes for Tony's team and students. ### AI-Powered Grading - Over 2,000 student submissions processed - 30,000+ problems auto-graded - 20,000+ high-confidence results (~95% accuracy) - More than 150 cumulative hours saved in TA effort per term (~250 student course) > Before Pensive, giving feedback individually to 300 students was impossible. Now, it's actually possible to give every student customized feedback. > > — Tony Dear ### AI Tutoring Tony also integrated Pensive's AI Tutor into his course, letting students solve problems and chat with the AI tutor for hints. This improved engagement, especially among quieter students. Unlike TA’s and staff members, the AI Tutor was available to his students 24/7, including late nights when students were finishing assignments or prepping for exams. ## Comparison to Alternatives Prior to using Pensive, Tony had experimented with other grading platforms without AI, including Gradescope. While Gradescope had reduced the time they spent grading student work by helping organize the grading process, it never provided the AI-assistance that Tony saw with Pensive. > At first glance, Pensive seems similar to Gradescope. But once I started using the AI tools, I quickly discovered that it's a superior platform. > > — Tony Dear ## The Grader Experience TAs quickly adapted thanks to Pensive's similarity to existing tools. Features like AI rubric generation and auto-grading automation made the transition seamless and required minimal extra efforts from Tony to educate them. > TAs now spend less time grading and more time developing as educators. Reading AI feedback helps them learn how to communicate constructive feedback themselves. > > — Tony Dear Tony now sees TA’s being able to provide more detailed, personalized feedback, whether on student papers or during tutoring sessions, which they now have more time to provide. ## Conclusion Pensive has transformed the way Tony Dear manages his large computer science courses. By combining AI-assistance with personalization, Tony's team achieved greater efficiency, richer feedback loops, and more learning opportunities for his students. > Pensive benefits both educators and students. It saves time, enhances feedback, and enriches the learning experience without compromising academic integrity. > > — Tony Dear